sched/cache: Keep nr_pref_llc_running in the runnable domain, to fix LLC mis-scheduling buggrafted

alb_break_llc() decides whether to break LLC preference during active
load balance. It does so by testing that every runnable fair task on the
source rq prefers its LLC:

	env->src_rq->nr_pref_llc_running == env->src_rq->cfs.h_nr_runnable

But the two counters cover different sets. nr_pref_llc_running is updated
in account_llc_enqueue()/account_llc_dequeue(), next to cfs_rq->nr_queued,
so it follows queued tasks. h_nr_runnable is updated in set_delayed()/
clear_delayed() and drops delay-dequeued tasks.

So under DELAY_DEQUEUE, a preferring task that goes to sleep stays counted
in nr_pref_llc_running while h_nr_runnable falls. The equality then breaks,
alb_break_llc() returns false, and active balance is free to pull a task
off its preferred LLC. Active balance only moves runnable tasks, and this
is the only LLC check it consults: once the stopper runs, LBF_ACTIVE_LB
skips the per-task test in can_migrate_task(). The runnable set is the one
we want.

Fix it on the counter side. A task should be counted in
nr_pref_llc_running exactly while it is both queued on its preferred LLC
(pref_llc_queued) and runnable (!sched_delayed). Define that membership
once in task_pref_llc_runnable(), and adjust the counter only through
pref_llc_running_inc()/pref_llc_running_dec() from the four sites that
change either input: account_llc_enqueue(), account_llc_dequeue(),
set_delayed() and clear_delayed(). Gating every update on the same
predicate keeps the delay, wake and dequeue paths from double-counting
or underflowing; see the comments at those sites for the ordering.

nr_llc_running and sd->llc_counts are not touched and stay on queued
semantics.

Fixes: 714059f79ff0 ("sched/cache: Handle moving single tasks to/from their preferred LLC")
Closes: https://lore.kernel.org/lkml/20260827135000.735138-1-zhanxusheng@xiaomi.com/
Reported-by: Zhan Xusheng <zhanxusheng@xiaomi.com>
Suggested-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Kayra Cizmeci <kayracizmeci@gmail.com>
Cc: <stable@kernel.org> # v7.2.x
Link: https://patch.msgid.link/06af61afedac32e6477f57feb4d658f6c411c3af.1790035273.git.tim.c.chen@linux.intel.com

1 files changed, 104 insertions, 0 deletions

diff --git a/drivers/scsi/arm/queue.h b/drivers/scsi/arm/queue.h
new file mode 100644
index 000000000..cb51379dc
--- /dev/null
+++ b/drivers/scsi/arm/queue.h
@@ -0,0 +1,104 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ *  linux/drivers/acorn/scsi/queue.h: queue handling
+ *
+ *  Copyright (C) 1997 Russell King
+ */
+#ifndef QUEUE_H
+#define QUEUE_H
+
+typedef struct {
+	struct list_head head;
+	struct list_head free;
+	spinlock_t queue_lock;
+	void *alloc;			/* start of allocated mem */
+} Queue_t;
+
+/*
+ * Function: void queue_initialise (Queue_t *queue)
+ * Purpose : initialise a queue
+ * Params  : queue - queue to initialise
+ */
+extern int queue_initialise (Queue_t *queue);
+
+/*
+ * Function: void queue_free (Queue_t *queue)
+ * Purpose : free a queue
+ * Params  : queue - queue to free
+ */
+extern void queue_free (Queue_t *queue);
+
+/*
+ * Function: struct scsi_cmnd *queue_remove (queue)
+ * Purpose : removes first SCSI command from a queue
+ * Params  : queue   - queue to remove command from
+ * Returns : struct scsi_cmnd if successful (and a reference), or NULL if no command available
+ */
+extern struct scsi_cmnd *queue_remove (Queue_t *queue);
+
+/*
+ * Function: struct scsi_cmnd *queue_remove_exclude_ref (queue, exclude)
+ * Purpose : remove a SCSI command from a queue
+ * Params  : queue   - queue to remove command from
+ *	     exclude - array of busy LUNs
+ * Returns : struct scsi_cmnd if successful (and a reference), or NULL if no command available
+ */
+extern struct scsi_cmnd *queue_remove_exclude(Queue_t *queue,
+					      unsigned long *exclude);
+
+#define queue_add_cmd_ordered(queue,SCpnt) \
+	__queue_add(queue,SCpnt,(SCpnt)->cmnd[0] == REQUEST_SENSE)
+#define queue_add_cmd_tail(queue,SCpnt) \
+	__queue_add(queue,SCpnt,0)
+/*
+ * Function: int __queue_add(Queue_t *queue, struct scsi_cmnd *SCpnt, int head)
+ * Purpose : Add a new command onto a queue
+ * Params  : queue - destination queue
+ *	     SCpnt - command to add
+ *	     head  - add command to head of queue
+ * Returns : 0 on error, !0 on success
+ */
+extern int __queue_add(Queue_t *queue, struct scsi_cmnd *SCpnt, int head);
+
+/*
+ * Function: struct scsi_cmnd *queue_remove_tgtluntag (queue, target, lun, tag)
+ * Purpose : remove a SCSI command from the queue for a specified target/lun/tag
+ * Params  : queue  - queue to remove command from
+ *	     target - target that we want
+ *	     lun    - lun on device
+ *	     tag    - tag on device
+ * Returns : struct scsi_cmnd if successful, or NULL if no command satisfies requirements
+ */
+extern struct scsi_cmnd *queue_remove_tgtluntag(Queue_t *queue, int target,
+						int lun, int tag);
+
+/*
+ * Function: queue_remove_all_target(queue, target)
+ * Purpose : remove all SCSI commands from the queue for a specified target
+ * Params  : queue  - queue to remove command from
+ *           target - target device id
+ * Returns : nothing
+ */
+extern void queue_remove_all_target(Queue_t *queue, int target);
+
+/*
+ * Function: int queue_probetgtlun (queue, target, lun)
+ * Purpose : check to see if we have a command in the queue for the specified
+ *	     target/lun.
+ * Params  : queue  - queue to look in
+ *	     target - target we want to probe
+ *	     lun    - lun on target
+ * Returns : 0 if not found, != 0 if found
+ */
+extern int queue_probetgtlun (Queue_t *queue, int target, int lun);
+
+/*
+ * Function: int queue_remove_cmd (Queue_t *queue, struct scsi_cmnd *SCpnt)
+ * Purpose : remove a specific command from the queues
+ * Params  : queue - queue to look in
+ *	     SCpnt - command to find
+ * Returns : 0 if not found
+ */
+int queue_remove_cmd(Queue_t *queue, struct scsi_cmnd *SCpnt);
+
+#endif /* QUEUE_H */